Explainable Predictive Models in Learning Analytics: A Proposal for Identifying Performance Patterns at UNIVESP
Year: 2026
Type: conference-paper
Abstract: This paper proposes an explainable Learning Analytics (LA) model to identify performance patterns and dropout risks at UNIVESP. It integrates Virtual Learning Environment (VLE) logs and social media textual data, processing them via Explainable Artificial Intelligence (XAI). Aiming at an institution with over 90,000 students, the model incorporates social variables in strict adherence to data privacy (LGPD). By using Gradient Boosting and SHAP values, the approach overcomes the accuracy-interpretability trade-off, mitigating algorithmic aversion. This ensures predictions remain pedagogically actionable, fostering personalized interventions and transforming distance education management on a large scale.
Institution: Universidade Virtual do Estado de São Paulo
Language: English
Subfield: Computer Science Applications
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7223119341
